arXiv subjects

Yao Liu

Publications and source records attributed to Yao Liu.

At least 145 records · Page 8Linked to original sources

arXiv2022

Offline Policy Optimization with Eligible Actions

Offline policy optimization could have a large impact on many real-world decision-making problems, as online learning may be infeasible in many applications. Importance sampling and its variants are a commonly used type of estimator in offline policy evaluation, and such estimators typically do not require assumptions on the properties and representational capabilities of value function or decision process model function classes. In this paper, we identify an important overfitting phenomenon in optimizing the importance weighted return, in which it may be possible for the learned policy to essentially avoid making aligned decisions for part of the initial state space. We propose an algorithm to avoid this overfitting through a new per-state-neighborhood normalization constraint, and provide a theoretical justification of the proposed algorithm. We also show the limitations of previous attempts to this approach. We test our algorithm in a healthcare-inspired simulator, a logged dataset collected from real hospitals and continuous control tasks. These experiments show the proposed method yields less overfitting and better test performance compared to state-of-the-art batch reinforcement learning algorithms.

cs.LG↗

arXiv2022

Gas Column Density Distribution of Molecular Clouds in the Third Quadrant of the Milky Way

We have obtained column density maps for an unbiased sample of 120 molecular clouds in the third quadrant of the Milky Way mid-plane (b$\le |5|^{\circ}$) within the galactic longitude range from 195$^{\circ}$ to 225$^{\circ}$, using the high sensitivity $^{12}$CO and $^{13}$CO ($J=1-0$) data from the Milky Way Imaging Scroll Painting (MWISP) project. The probability density functions of the molecular hydrogen column density of the clouds, N-PDFs, are fitted with both log-normal (LN) function and log-normal plus power-law (LN+PL) function. The molecular clouds are classified into three categories according to their shapes of N-PDFs, i.e., LN, LN+PL, and UN (unclear), respectively. About 72\% of the molecular clouds fall into the LN category, while 18\% and 10\% into the LN+PL and UN categories, respectively. A power-law scaling relation, $σ_s\propto N_{H_2}^{0.44}$, exists between the width of the N-PDF, $σ_s$, and the average column density, $N_{H_2}$, of the molecular clouds. However, $σ_s$ shows no correlation with the mass of the clouds. A correlation is found between the dispersion of normalized column density, $σ_{N/\rm }$, and the sonic Mach number, $\mathcal{M}$, of molecular clouds. Overall, as predicted by numerical simulations, the N-PDFs of the molecular clouds with active star formation activity tend to have N-PDFs with power-law high-density tails.

astro-ph.GA↗

arXiv2022

Generalized Federated Learning via Sharpness Aware Minimization

Federated Learning (FL) is a promising framework for performing privacy-preserving, distributed learning with a set of clients. However, the data distribution among clients often exhibits non-IID, i.e., distribution shift, which makes efficient optimization difficult. To tackle this problem, many FL algorithms focus on mitigating the effects of data heterogeneity across clients by increasing the performance of the global model. However, almost all algorithms leverage Empirical Risk Minimization (ERM) to be the local optimizer, which is easy to make the global model fall into a sharp valley and increase a large deviation of parts of local clients. Therefore, in this paper, we revisit the solutions to the distribution shift problem in FL with a focus on local learning generality. To this end, we propose a general, effective algorithm, \texttt{FedSAM}, based on Sharpness Aware Minimization (SAM) local optimizer, and develop a momentum FL algorithm to bridge local and global models, \texttt{MoFedSAM}. Theoretically, we show the convergence analysis of these two algorithms and demonstrate the generalization bound of \texttt{FedSAM}. Empirically, our proposed algorithms substantially outperform existing FL studies and significantly decrease the learning deviation.

cs.LG↗

arXiv2022

Provably Sample-Efficient RL with Side Information about Latent Dynamics

We study reinforcement learning (RL) in settings where observations are high-dimensional, but where an RL agent has access to abstract knowledge about the structure of the state space, as is the case, for example, when a robot is tasked to go to a specific room in a building using observations from its own camera, while having access to the floor plan. We formalize this setting as transfer reinforcement learning from an abstract simulator, which we assume is deterministic (such as a simple model of moving around the floor plan), but which is only required to capture the target domain's latent-state dynamics approximately up to unknown (bounded) perturbations (to account for environment stochasticity). Crucially, we assume no prior knowledge about the structure of observations in the target domain except that they can be used to identify the latent states (but the decoding map is unknown). Under these assumptions, we present an algorithm, called TASID, that learns a robust policy in the target domain, with sample complexity that is polynomial in the horizon, and independent of the number of states, which is not possible without access to some prior knowledge. In synthetic experiments, we verify various properties of our algorithm and show that it empirically outperforms transfer RL algorithms that require access to "full simulators" (i.e., those that also simulate observations).

cs.LG↗

arXiv2022

The Ages of Optically Bright Sub-Clusters in the Serpens Star-Forming Region

The Serpens Molecular Cloud is one of the most active star-forming regions within 500 pc, with over one thousand of YSOs at different evolutionary stages. The ages of the member stars inform us about the star formation history of the cloud. In this paper, we develop a spectral energy distribution (SED) fitting method for nearby evolved (diskless) young stars from members of the Pleiades to estimate their ages, with a temperature scale adopted from APOGEE spectra. When compared with literature temperatures of selected YSOs in Orion, the SED fits to cool (<5000 K) stars have temperatures that differ by an average of <~ 50 K and have a scatter of ~ 210 K for both disk-hosting and diskless stars. We then apply this method to YSOs in the Serpens Molecular Cloud to estimate ages of optical members previously identified from Gaia DR2 astrometry data. The optical members in Serpens are concentrated in different subgroups with ages from ~4 Myr to ~22 Myr; the youngest clusters, W40 and Serpens South, are dusty regions that lack enough optical members to be included in this analysis. These ages establish that the Serpens Molecular Cloud has been forming stars for much longer than has been inferred from infrared surveys.

astro-ph.SR↗

arXiv2022

Molecules with ALMA at Planet-forming Scales (MAPS) III: Characteristics of Radial Chemical Substructures

The Molecules with ALMA at Planet-forming Scales (MAPS) Large Program provides a detailed, high resolution (${\sim}$10-20 au) view of molecular line emission in five protoplanetary disks at spatial scales relevant for planet formation. Here, we present a systematic analysis of chemical substructures in 18 molecular lines toward the MAPS sources: IM Lup, GM Aur, AS 209, HD 163296, and MWC 480. We identify more than 200 chemical substructures, which are found at nearly all radii where line emission is detected. A wide diversity of radial morphologies - including rings, gaps, and plateaus - is observed both within each disk and across the MAPS sample. This diversity in line emission profiles is also present in the innermost 50 au. Overall, this suggests that planets form in varied chemical environments both across disks and at different radii within the same disk. Interior to 150 au, the majority of chemical substructures across the MAPS disks are spatially coincident with substructures in the millimeter continuum, indicative of physical and chemical links between the disk midplane and warm, elevated molecular emission layers. Some chemical substructures in the inner disk and most chemical substructures exterior to 150 au cannot be directly linked to dust substructure, however, which indicates that there are also other causes of chemical substructures, such as snowlines, gradients in UV photon fluxes, ionization, and radially-varying elemental ratios. This implies that chemical substructures could be developed into powerful probes of different disk characteristics, in addition to influencing the environments within which planets assemble. This paper is part of the MAPS special issue of the Astrophysical Journal Supplement.

astro-ph.EP↗

arXiv2022

Information-Theoretic Limits of Integrated Sensing and Communication with Correlated Sensing and Channel States for Vehicular Networks

In connected vehicular networks, it is vital to have vehicular nodes that are capable of sensing about surrounding environments and exchanging messages with each other for automating and coordinating purpose. Towards this end, integrated sensing and communication (ISAC), combining both sensing and communication systems to jointly utilize their resources and to pursue mutual benefits, emerges as a new cost-effective solution. In ISAC, the hardware and spectrum co-sharing leads to a fundamental tradeoff between sensing and communication performance, which is not well understood except for very simple cases with the same sensing and channel states, and perfect channel state information at the receiver (CSIR). In this paper, a general point-to-point ISAC model is proposed to account for the scenarios that the sensing state is different from but correlated with the channel state, and the CSIR is not necessarily perfect. For the model considered, the optimal tradeoff is characterized by a capacity-distortion function that quantifies the best communication rate for a given sensing distortion constraint requirement. An iterative algorithm is proposed to compute such tradeoff, and a few non-trivial examples are constructed to demonstrate the benefits of ISAC as compared to the separation-based approach.

cs.IT↗

arXiv2022

Modeling the arc and ring structures in the HD 143006 disk

Rings and asymmetries in protoplanetary disks are considered as signposts of ongoing planet formation. In this work, we conduct three-dimensional radiative transfer simulations to model the intriguing disk around HD 143006 that has three dust rings and a bright arc. A complex geometric configuration, with a misaligned inner disk, is assumed to account for the asymmetric structures. The two-dimensional surface density is constructed by iteratively fitting the ALMA data. We find that the dust temperature displays a notable discontinuity at the boundary of the misalignment. The ring masses range from 0.6 to 16Mearth that are systematically lower than those inferred in the younger HL Tau disk. The arc occupies nearly 20% of the total dust mass. Such a high mass fraction of dust grains concentrated in a local region is consistent with the mechanism of dust trapping into vortices. Assuming a gas-to-dust mass ratio of 30 that is constant throughout the disk, the dense and cold arc is close to the threshold of being gravitationally unstable, with the Toomre parameter Q~1.3. Nevertheless, our estimate of Q relies on the assumption for the unknown gas-to-dust mass ratio. Adopting a lower gas-to-dust mass ratio would increase the inferred Q value. Follow-up high resolution observations of dust and gas lines are needed to clarify the origin of the substructures.

astro-ph.EP↗

arXiv2022

Learning to Guide Human Attention on Mobile Telepresence Robots with 360 Vision

Mobile telepresence robots (MTRs) allow people to navigate and interact with a remote environment that is in a place other than the person's true location. Thanks to the recent advances in 360 degree vision, many MTRs are now equipped with an all-degree visual perception capability. However, people's visual field horizontally spans only about 120 degree of the visual field captured by the robot. To bridge this observability gap toward human-MTR shared autonomy, we have developed a framework, called GHAL360, to enable the MTR to learn a goal-oriented policy from reinforcements for guiding human attention using visual indicators. Three telepresence environments were constructed using datasets that are extracted from Matterport3D and collected from a real robot respectively. Experimental results show that GHAL360 outperformed the baselines from the literature in the efficiency of a human-MTR team completing target search tasks.

cs.RO↗

arXiv2022

Tiny Object Tracking: A Large-scale Dataset and A Baseline

Tiny objects, frequently appearing in practical applications, have weak appearance and features, and receive increasing interests in meany vision tasks, such as object detection and segmentation. To promote the research and development of tiny object tracking, we create a large-scale video dataset, which contains 434 sequences with a total of more than 217K frames. Each frame is carefully annotated with a high-quality bounding box. In data creation, we take 12 challenge attributes into account to cover a broad range of viewpoints and scene complexities, and annotate these attributes for facilitating the attribute-based performance analysis. To provide a strong baseline in tiny object tracking, we propose a novel Multilevel Knowledge Distillation Network (MKDNet), which pursues three-level knowledge distillations in a unified framework to effectively enhance the feature representation, discrimination and localization abilities in tracking tiny objects. Extensive experiments are performed on the proposed dataset, and the results prove the superiority and effectiveness of MKDNet compared with state-of-the-art methods. The dataset, the algorithm code, and the evaluation code are available at https://github.com/mmic-lcl/Datasets-and-benchmark-code.

cs.CV↗

arXiv2022

LoMar: A Local Defense Against Poisoning Attack on Federated Learning

Federated learning (FL) provides a high efficient decentralized machine learning framework, where the training data remains distributed at remote clients in a network. Though FL enables a privacy-preserving mobile edge computing framework using IoT devices, recent studies have shown that this approach is susceptible to poisoning attacks from the side of remote clients. To address the poisoning attacks on FL, we provide a \textit{two-phase} defense algorithm called {Lo}cal {Ma}licious Facto{r} (LoMar). In phase I, LoMar scores model updates from each remote client by measuring the relative distribution over their neighbors using a kernel density estimation method. In phase II, an optimal threshold is approximated to distinguish malicious and clean updates from a statistical perspective. Comprehensive experiments on four real-world datasets have been conducted, and the experimental results show that our defense strategy can effectively protect the FL system. {Specifically, the defense performance on Amazon dataset under a label-flipping attack indicates that, compared with FG+Krum, LoMar increases the target label testing accuracy from $96.0\%$ to $98.8\%$, and the overall averaged testing accuracy from $90.1\%$ to $97.0\%$.

cs.LG↗

arXiv2021

Context-Aware Online Client Selection for Hierarchical Federated Learning

Federated Learning (FL) has been considered as an appealing framework to tackle data privacy issues of mobile devices compared to conventional Machine Learning (ML). Using Edge Servers (ESs) as intermediaries to perform model aggregation in proximity can reduce the transmission overhead, and it enables great potentials in low-latency FL, where the hierarchical architecture of FL (HFL) has been attracted more attention. Designing a proper client selection policy can significantly improve training performance, and it has been extensively used in FL studies. However, to the best of our knowledge, there are no studies focusing on HFL. In addition, client selection for HFL faces more challenges than conventional FL, e.g., the time-varying connection of client-ES pairs and the limited budget of the Network Operator (NO). In this paper, we investigate a client selection problem for HFL, where the NO learns the number of successful participating clients to improve the training performance (i.e., select as many clients in each round) as well as under the limited budget on each ES. An online policy, called Context-aware Online Client Selection (COCS), is developed based on Contextual Combinatorial Multi-Armed Bandit (CC-MAB). COCS observes the side-information (context) of local computing and transmission of client-ES pairs and makes client selection decisions to maximize NO's utility given a limited budget. Theoretically, COCS achieves a sublinear regret compared to an Oracle policy on both strongly convex and non-convex HFL. Simulation results also support the efficiency of the proposed COCS policy on real-world datasets.

cs.LG↗

arXiv2021

Molecules with ALMA at Planet-forming Scales (MAPS). X. Studying deuteration at high angular resolution toward protoplanetary disks

Deuterium fractionation is dependent on various physical and chemical parameters. Thus, the formation location and thermal history of material in the solar system is often studied by measuring its D/H ratio. This requires knowledge about the deuteration processes operating during the planet formation era. We aim to study these processes by radially resolving the DCN/HCN (at 0.3" resolution) and N$_2$D$^+$/N$_2$H$^+$ (0.3 to 0.9") column density ratios toward the five protoplanetary disks observed by the Molecules with ALMA at Planet-forming scales (MAPS) Large Program. DCN is detected in all five sources, with one newly reported detection. N$_2$D$^+$ is detected in four sources, two of which are newly reported detections. We derive column density profiles that allow us to study the spatial variation of the DCN/HCN and N$_2$D$^+$/N$_2$H$^+$ ratios at high resolution. DCN/HCN varies considerably for different parts of the disks, ranging from $10^{-3}$ to $10^{-1}$. In particular, the inner disk regions generally show significantly lower HCN deuteration compared with the outer disk. In addition, our analysis confirms that two deuterium fractionation channels are active, which can alter the D/H ratio within the pool of organic molecules. N$_2$D$^+$ is found in the cold outer regions beyond $\sim$50 au, with N$_2$D$^+$/N$_2$H$^+$ ranging between $10^{-2}$ and 1 across the disk sample. This is consistent with the theoretical expectation that N$_2$H$^+$ deuteration proceeds via the low-temperature channel only. This paper is part of the MAPS special issue of the Astrophysical Journal Supplement.

astro-ph.SR↗

arXiv2021

Molecules with ALMA at Planet-forming Scales (MAPS) V: CO gas distributions

Here we present high resolution (15-24 au) observations of CO isotopologue lines from the Molecules with ALMA on Planet-forming Scales (MAPS) ALMA Large Program. Our analysis employs $^{13}$CO and C$^{18}$O ($J$=2-1), (1-0), and C$^{17}$O (1-0) line observations of five protoplanetary disks. We retrieve CO gas density distributions, using three independent methods: (1) a thermo-chemical modeling framework based on the CO data, the broadband spectral energy distribution, and the mm-continuum emission; (2) an empirical temperature distribution based on optically thick CO lines; and (3) a direct fit to the C$^{17}$O hyperfine lines. Results from these methods generally show excellent agreement. The CO gas column density profiles of the five disks show significant variations in the absolute value and the radial shape. Assuming a gas-to-dust mass ratio of 100, all five disks have a global CO-to-H$_2$ abundance of 10-100 times lower than the ISM ratio. The CO gas distributions between 150-400 au match well with models of viscous disks, supporting the long-standing theory. CO gas gaps appear to be correlated with continuum gap locations, but some deep continuum gaps do not have corresponding CO gaps. The relative depths of CO and dust gaps are generally consistent with predictions of planet-disk interactions, but some CO gaps are 5-10 times shallower than predictions based on dust gaps. This paper is part of the MAPS special issue of the Astrophysical Journal Supplement.

astro-ph.EP↗

arXiv2021

Molecules with ALMA at Planet-forming Scales (MAPS) XIV: Revealing disk substructures in multi-wavelength continuum emission

Constraining dust properties of planet-forming disks via high angular resolution observations is fundamental to understanding how solids are trapped in substructures and how dust growth may be favored or accelerated therein. We use ALMA dust continuum observations of the Molecules with ALMA at Planet-forming Scales (MAPS) disks and explore a large parameter space to constrain the radial distribution of solid mass and maximum grain size in each disk, including or excluding dust scattering. In the nonscattering model, the dust surface density and maximum grain size profiles decrease from the inner disks to the outer disks, with local maxima at the bright ring locations, as expected from dust trapping models. The inferred maximum grain sizes from the inner to outer disks decrease from ~1 cm to 1 mm. For IM Lup, HD 163296, and MWC 480 in the scattering model, two solutions are compatible with their observed inner disk emission: one solution corresponding to a maximum grain size of a few millimeters (similar to the nonscattering model), and the other corresponding to a few hundred micrometer sizes. Based on the estimated Toomre parameter, only IM Lup -- which shows a prominent spiral morphology in millimeter dust -- is found to be gravitationally unstable. The estimated maximum Stokes number in all the disks lies between 0.01 and 0.3, and the estimated turbulence parameters in the rings of AS 209 and HD 163296 are close to the threshold where dust growth is limited by turbulent fragmentation. This paper is part of the MAPS special issue of the Astrophysical Journal Supplement.

astro-ph.EP↗

arXiv2021

Molecules with ALMA at Planet-forming Scales (MAPS) IV: Emission Surfaces and Vertical Distribution of Molecules

The Molecules with ALMA at Planet-forming Scales (MAPS) Large Program provides a unique opportunity to study the vertical distribution of gas, chemistry, and temperature in the protoplanetary disks around IM Lup, GM Aur, AS 209, HD 163296, and MWC 480. By using the asymmetry of molecular line emission relative to the disk major axis, we infer the emission height ($z$) above the midplane as a function of radius ($r$). Using this method, we measure emitting surfaces for a suite of CO isotopologues, HCN, and C$_2$H. We find that $^{12}$CO emission traces the most elevated regions with $z/r > 0.3$, while emission from the less abundant $^{13}$CO and C$^{18}$O probes deeper into the disk at altitudes of $z/r \lesssim 0.2$. C$_2$H and HCN have lower opacities and SNRs, making surface fitting more difficult, and could only be reliably constrained in AS 209, HD 163296, and MWC 480, with $z/r \lesssim 0.1$, i.e., relatively close to the planet-forming midplanes. We determine peak brightness temperatures of the optically thick CO isotopologues and use these to trace 2D disk temperature structures. Several CO temperature profiles and emission surfaces show dips in temperature or vertical height, some of which are associated with gaps and rings in line and/or continuum emission. These substructures may be due to local changes in CO column density, gas surface density, or gas temperatures, and detailed thermo-chemical models are necessary to better constrain their origins and relate the chemical compositions of elevated disk layers with those of planet-forming material in disk midplanes. This paper is part of the MAPS special issue of the Astrophysical Journal Supplement.

astro-ph.GA↗

arXiv2021

Molecules with ALMA at Planet-forming Scales (MAPS) VIII: CO Gap in AS 209--Gas Depletion or Chemical Processing?

Emission substructures in gas and dust are common in protoplanetary disks. Such substructures can be linked to planet formation or planets themselves. We explore the observed gas substructures in AS 209 using thermochemical modeling with RAC2D and high-spatial resolution data from the Molecules with ALMA at Planet-forming Scales(MAPS) program. The observations of C$^{18}$O J=2-1 emission exhibit a strong depression at 88 au overlapping with the positions of multiple gaps in millimeter dust continuum emission. We find that the observed CO column density is consistent with either gas surface-density perturbations or chemical processing, while C$_2$H column density traces changes in the C/O ratio rather than the H$_2$ gas surface density. However, the presence of a massive planet (> 0.2 M$_{Jup}$) would be required to account for this level of gas depression, which conflicts with constraints set by the dust emission and the pressure profile measured by gas kinematics. Based on our models, we infer that a local decrease of CO abundance is required to explain the observed structure in CO, dominating over a possible gap-carving planet present and its effect on the H$_2$ surface density. This paper is part of the MAPS special issue of the Astrophysical Journal Supplement Series.

astro-ph.EP↗

arXiv2021

Molecules with ALMA at Planet-forming Scales (MAPS) I: Program Overview and Highlights

Planets form and obtain their compositions in dust and gas-rich disks around young stars, and the outcome of this process is intimately linked to the disk chemical properties. The distributions of molecules across disks regulate the elemental compositions of planets, including C/N/O/S ratios and metallicity (O/H and C/H), as well as access to water and prebiotically relevant organics. Emission from molecules also encodes information on disk ionization levels, temperature structures, kinematics, and gas surface densities, which are all key ingredients of disk evolution and planet formation models. The Molecules with ALMA at Planet-forming Scales (MAPS) ALMA Large Program was designed to expand our understanding of the chemistry of planet formation by exploring disk chemical structures down to 10 au scales. The MAPS program focuses on five disks - around IM Lup, GM Aur, AS 209, HD 163296, and MWC 480 - in which dust substructures are detected and planet formation appears to be ongoing. We observed these disks in 4 spectral setups, which together cover ~50 lines from over 20 different species. This paper introduces the ApJS MAPS Special Issue by presenting an overview of the program motivation, disk sample, observational details, and calibration strategy. We also highlight key results, including discoveries of links between dust, gas, and chemical sub-structures, large reservoirs of nitriles and other organics in the inner disk regions, and elevated C/O ratios across most disks. We discuss how this collection of results is reshaping our view of the chemistry of planet formation.

astro-ph.EP↗